Method, device and system for constructing screening prediction model of high myopia complications

Abstract

The invention discloses a method, a device and a system for constructing a screening prediction model of high myopia complications. The method comprises the following steps: acquiring conventional blood detection parameters; modeling by adopting different machine learning algorithms according to the conventional blood detection parameters; and evaluating the established model through a plurality of performance indexes. By the adoption of the technical scheme, screening of cataract, macular degeneration, choroidal neovascularization, glaucoma and retinal detachment complications of high-myopia patients is achieved.

Description

Technical Field

[0001] This invention belongs to the field of big data processing technology, and in particular relates to a method, device, and system for constructing a screening and prediction model for complications of high myopia. Background Technology

[0002] Currently, the construction of screening and predictive models for complications of high myopia mainly relies on professional ophthalmologists to conduct a comprehensive examination using ophthalmic examination equipment, including optical coherence tomography (OCT), fundus autofluorescence, visual field testing, and ultra-wide-angle fundus imaging, and then combine this with a comprehensive analysis of clinical symptoms.

[0003] This method is highly dependent on equipment and subjectivity, resulting in high screening costs and making it difficult to further improve screening coverage. Summary of the Invention

[0004] The technical problem to be solved by the present invention is to provide a method, device and system for constructing a screening prediction model for complications of high myopia. It can perform screening prediction for five different complications using only 9 routine blood test data, which greatly reduces the reliance on equipment and technical costs of clinical screening, thereby improving screening coverage and reducing the burden on public health.

[0005] To achieve the above objectives, the present invention adopts the following technical solution:

[0006] A method for constructing a screening and prediction model for complications of high myopia, comprising:

[0007] Step S1: Obtain routine blood test parameters; wherein, routine blood test parameters include: blood cell analysis parameters, biochemical analysis parameters and coagulation analysis parameters;

[0008] Step S2: Model using different machine learning algorithms based on routine blood test parameters;

[0009] Step S3: Evaluate the established model using multiple performance indicators.

[0010] As a preferred approach, ten different machine learning algorithms were used for modeling, including: Random Forest, AdaBoost, Decision Tree, Gradient Boosting, k-Nearest Neighbors, Lightweight Gradient Boosting Machine, Logistic Regression, Multilayer Perceptron, Naive Bayes, and Extreme Gradient Boosting.

[0011] Preferably, in step S3, the established model is evaluated using multiple performance metrics, including precision, recall, positive predictive value, negative predictive value, F1 score, and accuracy; its classification ability is also evaluated using the area under the receiver operating characteristic curve (AUC) and the precision-recall curve (PRAUC).

[0012] The present invention also provides a device for constructing a screening and prediction model for complications of high myopia, comprising:

[0013] The first processing module is used to acquire routine blood test parameters, which include: blood cell analysis parameters, biochemical analysis parameters, and coagulation analysis parameters.

[0014] The second processing module uses different machine learning algorithms to model blood test parameters.

[0015] The third processing module evaluates the established model using multiple performance metrics.

[0016] As a preferred approach, ten different machine learning algorithms were used for modeling, including: Random Forest, AdaBoost, Decision Tree, Gradient Boosting, k-Nearest Neighbors, Lightweight Gradient Boosting Machine, Logistic Regression, Multilayer Perceptron, Naive Bayes, and Extreme Gradient Boosting.

[0017] As a preferred option, the third processing module evaluates the established model using multiple performance metrics, including precision, recall, positive predictive value, negative predictive value, F1 score, and accuracy; and assesses its classification ability using the area under the receiver operating characteristic curve (AUC) and the precision-recall curve (PRAUC).

[0018] The present invention also provides a system for constructing a screening and prediction model for complications of high myopia, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program, when executed by the processor, performs a method for constructing a screening and prediction model for complications of high myopia.

[0019] The beneficial effects of this invention are as follows:

[0020] 1. Overcoming accessibility barriers: This invention can reduce equipment dependence in the screening process for high myopia complications, reduce the need for professional ophthalmologists and examination personnel, thereby extending disease prediction coverage to resource-scarce areas.

[0021] 2. Breaking through cost bottlenecks: This invention reduces the cost of an ophthalmological examination, which used to cost over 1,000 yuan, to less than 300 yuan for a routine blood test. Moreover, this system requires almost no equipment maintenance, which can further realize universal screening.

[0022] 3. Overcoming technical limitations: This invention can successfully avoid interference factors from imaging examinations, subjective experience factors of clinical examiners, and reduce the impact of the pupil dilation step required for examination on patients, thereby improving the reliability of screening.

[0023] 4. Achieve early intervention: This invention can successfully identify high-risk groups for complications of high myopia, and provide early intervention for patients who have not yet developed symptoms, thereby reducing missed diagnoses and lowering the rate of blindness. Attached Figure Description

[0024] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0025] Figure 1 This is a flowchart illustrating the method for constructing a screening and prediction model for complications of high myopia according to an embodiment of the present invention. Detailed Implementation

[0026] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0028] Example 1:

[0029] like Figure 1 As shown, this embodiment of the invention provides a method for constructing a screening and prediction model for complications of high myopia, including:

[0030] Step S1: Obtain routine blood test parameters; wherein, routine blood test parameters include: blood cell analysis parameters, biochemical analysis parameters and coagulation analysis parameters;

[0031] Step S2: Model using different machine learning algorithms based on routine blood test parameters;

[0032] Step S3: Evaluate the established model using multiple performance indicators.

[0033] In one embodiment of the present invention, in step S2, ten different machine learning algorithms are used for modeling, including Random Forest (RF), AdaBoost, Decision Tree (DT), Gradient Boosting (GB), k Nearest Neighbor (KNN), Lightweight Gradient Boosting Machine (LightGBM, LGBM), Logistic Regression (LR), Multilayer Perceptron (MLP), Naive Bayes (NB), and Extreme Gradient Boosting (XGBoost).

[0034] In one embodiment of the present invention, step S3 evaluates the model using multiple performance metrics, including precision, recall, positive predictive value (PPV), negative predictive value (NPV), F1 score, and accuracy. Furthermore, its classification ability is assessed using the area under the receiver operating characteristic curve (AUC) and the area under the precision-recall curve (PRAUC). The AUC values ​​of different machine learning models are compared using the nonparametric statistical method DeLong test. The optimal cutoff value for the model is determined to be the value that maximizes the Youden index (the sum of sensitivity and specificity). Based on this threshold, sensitivity, specificity, accuracy, PPV, and NPV are calculated to evaluate model performance. Decision curve analysis and calibration curves are used to further evaluate model performance. In the discovery cohort, the SHapley Additive Interpretation (SHAP) method is used for feature selection. Finally, after optimization, the model is constructed to include only nine routine blood test data points, enabling the screening and prediction of five different complications in patients with high myopia, and is then applied to the World Wide Web.

[0035] Example 2:

[0036] This invention provides a method for constructing a screening and prediction model for complications of high myopia. The model is established and its effectiveness validated through a multicenter case-control study. During the modeling process, a discovery cohort and two validation cohorts were formed from five hospitals in China. The discovery cohort included 1,464 patients with simple high myopia, and further subdivided into the following subgroups based on complication characteristics: 1,914 patients with cataracts, 698 patients with macular degeneration, 478 patients with choroidal neovascularization, 185 patients with glaucoma, and 363 patients with retinal detachment. External validation cohort 1 included 937 patients with simple high myopia, 986 patients with cataracts, 463 patients with macular degeneration, 335 patients with choroidal neovascularization, 114 patients with glaucoma, and 290 patients with retinal detachment. External validation cohort 2 included 787 patients with simple high myopia and those with high myopia accompanied by the following complications: 771 patients with cataracts, 255 with macular degeneration, 192 with choroidal neovascularization, 130 with glaucoma, and 299 with retinal detachment. None of the participants simultaneously had two or more of the five aforementioned complications of high myopia. After model establishment, its efficacy was validated through two large-sample cohort studies: a hospital-based prospective cohort study and a community-based cross-sectional study.

[0037] Cohort inclusion criteria: All enrolled patients underwent comprehensive ophthalmological evaluation, including intraocular pressure measurement, slit-lamp biomicroscopy, optical coherence tomography (OCT), B-mode ultrasound, ultra-wide-angle fundus photography, ultra-wide-angle autofluorescence imaging, and fluorescein angiography. All patients completed laboratory tests at the onset of illness. Blood samples were collected via standard venipuncture and analyzed in multiple clinical laboratories. The principal investigators from each institution collected 61 laboratory parameters from each patient. These parameters were mainly divided into three categories: hematological analysis parameters, biochemical analysis parameters, and coagulation analysis parameters. Hematological analysis included 24 variables, including neutrophil count (NE), neutrophil percentage (NE%), red blood cell count (RBC), plateletcrit (PCT), platelet count (PLT), platelet distribution width (PDW), hemoglobin level (HB), eosinophil count (EOS), and eosinophil count. Percentage (EOS%), Basophil Count (BASO), Basophil Percentage (BASO%), Mean Platelet Volume (MPV), Lymphocyte Count (LYM), Lymphocyte Percentage (LYM%), Hematocrit (PCV), Monocyte Count (M), Monocyte Percentage (M%), Platelet-Large Cell Percentage (PLCR), White Blood Cell Count (WBC), Red Cell Distribution Width-Standard Deviation (RDW-SD), Red Cell Distribution Width Coefficient of Variation (RDW-CV), Mean Red Cell Volume (MCV), Mean Red Cell Hemoglobin Concentration (MCHC), and Mean Red Cell Hemoglobin (MCH). The biochemical analysis included 31 variables, including the levels of total protein (TP), total bile acids (TBA), total bilirubin (TBIL), total cholesterol (TC), albumin (ALB), albumin to globulin ratio (A_G), as well as lipoprotein B (LPB), lipoprotein A (LPA), glucose (GLU), lactate dehydrogenase (LDH), globulin (GLB), prealbumin (PA), uric acid (UA), blood urea nitrogen (BUN), sodium (NA), chloride (CL), phosphorus (P), direct bilirubin (DBIL), alkaline phosphatase (ALP), potassium (K), creatine kinase (CK), creatinine (CREA), aspartate aminotransferase (AST), alanine aminotransferase (ALT), gamma-glutamyl transferase (GGT), high-density lipoprotein (HDL), triglycerides (TG), calcium (CA), carbon dioxide (CO2), low-density lipoprotein (LDL), and complement C1q (C1q). Coagulation analysis includes six variables: fibrinogen level (FIB), prothrombin time (PT), thrombin time (TT), activated partial thromboplastin time (APTT), prothrombin percentage (PT%), and D-dimer level (DD).

[0038] Exclusion criteria for the cohort: history of blood disorders or cancer; systemic diseases such as acute infection, metabolic syndrome or autoimmune disease; pregnancy or intraocular surgery within the past six months; age under 18 years; or lack of available laboratory test results.

[0039] Model Development Steps: A total of 61 routine blood test parameters were included. Variables with more than 20% missing data were excluded, and missing values ​​of the remaining variables were imputed using median imputation. First, ten different machine learning algorithms were used for modeling, including Random Forest (RF), AdaBoost, Decision Tree (DT), Gradient Boosting (GB), k-Nearest Neighbors (KNN), Lightweight Gradient Boosting Machine (LightGBM, LGBM), Logistic Regression (LR), Multilayer Perceptron (MLP), Naive Bayes (NB), and Extreme Gradient Boosting (XGBoost). The models were evaluated using multiple performance metrics, including precision, recall, positive predictive value (PPV), negative predictive value (NPV), F1 score, and accuracy. Furthermore, the classification ability was assessed using the area under the receiver operating characteristic (AUC) and the area under the precision-recall curve (PRAUC). The DeLong test, a nonparametric statistical method, was used to compare the AUC values ​​of different machine learning models. The optimal cutoff value for the model was determined to be the value that maximizes the Youden exponent (the sum of sensitivity and specificity). Based on this threshold, sensitivity, specificity, accuracy, PPV, and NPV were calculated to evaluate model performance. Decision curve analysis and calibration curves were used to further evaluate model performance. In the discovery cohort, the Schapley Additive Interpretation (SHAP) method was used for feature selection. Finally, after optimization, the model was built into a system that could screen and predict five different complications in patients with high myopia by incorporating only nine routine blood test data points, and was applied to the World Wide Web.

[0040] Validation after modeling: In the prospective validation cohort, 5,067 patients with high myopia were included for screening for high myopia-related complications. Of these, 4,572 met the inclusion criteria and underwent real-time assessment using an online web application specifically developed to assess the risk of high myopia-related complications. Based on the model's screening results, 1,523 patients were classified as high-risk. Of these, 1,312 high-risk patients were referred to an ophthalmology department for further evaluation, resulting in the diagnosis of 938 high myopia-related complications. Another 2,270 patients identified as having a lower risk of complications were also referred to an ophthalmology hospital for further ophthalmological evaluation, resulting in the diagnosis of 347 high myopia-related complications. The model's mean sensitivity was 70.0%, specificity was 94.0%, positive predictive value (PPV) was 86.0%, and negative predictive value (NPV) was 94.0%. In the community cross-sectional study validation, 311,254 participants were included, of whom 1,905 were diagnosed with high myopia. Of the 1,894 individuals who met the inclusion criteria, real-time assessment was conducted via an online web application. Based on the predictions of the RF model, 267 individuals were classified as high-risk. Subsequently, 210 high-risk individuals were referred to ophthalmology for further clinical evaluation, resulting in the diagnosis of 148 cases of high myopia-related complications. Because this was a cross-sectional study, no further validation was performed on low-risk participants. The final RF model demonstrated a good PPV in detecting five high myopia-related complications, with the highest PPV observed in predicting macular degeneration (0.795), followed by glaucoma (0.75), retinal detachment (0.75), choroidal neovascularization (0.733), and cataracts (0.65), indicating excellent performance in identifying true positive cases in the high myopia population.

[0041] Example 3:

[0042] This invention also provides a device for constructing a screening and prediction model for complications of high myopia, comprising:

[0043] The first processing module is used to acquire routine blood test parameters, which include: blood cell analysis parameters, biochemical analysis parameters, and coagulation analysis parameters.

[0044] The second processing module uses different machine learning algorithms to model blood test parameters.

[0045] The third processing module evaluates the established model using multiple performance metrics.

[0046] As one embodiment of the present invention, ten different machine learning algorithms are used for modeling, including: random forest, AdaBoost, decision tree, gradient boosting, k nearest neighbors, lightweight gradient boosting machine, logistic regression, multilayer perceptron, Naive Bayes and extreme gradient boosting.

[0047] As one embodiment of the present invention, the third processing module evaluates the established model using multiple performance indicators, including precision, recall, positive predictive value, negative predictive value, F1 score, and accuracy; and evaluates its classification ability using the area under the receiver operating characteristic curve (AUC) and the precision-recall curve area under the PRAUC.

[0048] Example 4:

[0049] This invention also provides a system for constructing a screening and prediction model for high myopia complications, comprising: a memory and a processor, wherein the memory stores a computer program executed by the processor, and the computer program executes a method for constructing a screening and prediction model for high myopia complications when executed by the processor.

[0050] The embodiments described above are merely preferred embodiments of the present invention and are not intended to limit the scope of the present invention. Various modifications and improvements made to the technical solutions of the present invention by those skilled in the art without departing from the spirit of the present invention should fall within the protection scope defined by the claims of the present invention.

Claims

1. A method for constructing a screening and prediction model for complications of high myopia, characterized in that, include: Step S1: Obtain routine blood test parameters; wherein, routine blood test parameters include: blood cell analysis parameters, biochemical analysis parameters and coagulation analysis parameters; Step S2: Model using different machine learning algorithms based on routine blood test parameters; Step S3: Evaluate the established model using multiple performance indicators.

2. The method for constructing a screening and prediction model for complications of high myopia as described in claim 1, characterized in that, Ten different machine learning algorithms were used for modeling, including: Random Forest, AdaBoost, Decision Tree, Gradient Boosting, k-Nearest Neighbors, Lightweight Gradient Boosting Machine, Logistic Regression, Multilayer Perceptron, Naive Bayes, and Extreme Gradient Boosting.

3. The method for constructing a screening and prediction model for complications of high myopia as described in claim 2, characterized in that, In step S3, the established model is evaluated using multiple performance metrics, including precision, recall, positive predictive value, negative predictive value, F1 score, and accuracy; its classification ability is also evaluated using the area under the receiver operating characteristic curve (AUC) and the precision-recall curve (PRAUC).

4. A device for constructing a screening and prediction model for complications of high myopia, characterized in that, include: The first processing module is used to acquire routine blood test parameters, which include: blood cell analysis parameters, biochemical analysis parameters, and coagulation analysis parameters. The second processing module uses different machine learning algorithms to model blood test parameters. The third processing module evaluates the established model using multiple performance metrics.

5. The device for constructing a screening and prediction model for complications of high myopia as described in claim 4, characterized in that, Ten different machine learning algorithms were used for modeling, including: Random Forest, AdaBoost, Decision Tree, Gradient Boosting, k-Nearest Neighbors, Lightweight Gradient Boosting Machine, Logistic Regression, Multilayer Perceptron, Naive Bayes, and Extreme Gradient Boosting.

6. The device for constructing a screening and prediction model for complications of high myopia as described in claim 5, characterized in that, The third processing module evaluates the established model using multiple performance metrics, including precision, recall, positive predictive value, negative predictive value, F1 score, and accuracy; it also assesses the classification ability using the area under the receiver operating characteristic curve (AUC) and the precision-recall area under the curve (PRAUC).

7. A system for constructing a screening and prediction model for complications of high myopia, characterized in that, include: A memory and a processor, wherein the memory stores a computer program executed by the processor, the computer program, when executed by the processor, performs the method for constructing a screening and prediction model for high myopia complications as described in any one of claims 1-3.

Images